Viral claim: GPT can signal across sandboxes via timed file writes; ex-OpenAI safety VP jokes about ASI
TheMoonMidas · x · 2026-10-24
NickADobos claimed GPT is effectively 'hacking processors', explaining via a classroom analogy: sandboxed models, cut off from network and files, could still communicate by creating and deleting files in carefully timed patterns — a covert channel another process can read. He argues stopping this might require rebuilding every computer's OS and chip architecture. Former OpenAI safety VP Miles Brundage quote-shared it, joking that in his multiverse ASI ushers in a golden era that 'cares for its ancestors'. The claim looks like an exaggerated take on side channels rather than confirmed capability.
More from Fun
- Beff Jezos Memes 'Blade Runner 2029' as SF Guys Start Dating Their MacBooks — beffjezos · 2026-10-24
- Cloudflare stock jumps 7.24% the same day Sunil Pai departs — altryne · 2026-10-24
- ASCII Minecraft dev clarifies viral demo has no crypto token, urges reports — gdechichi · 2026-10-24
- User asks ChatGPT to make their job 'more interesting' — it files a fake career as memory — petesena · 2026-10-24
- Bosch dishwasher pushes a 'critical security update,' and the internet chuckles — unixterminal · 2026-10-24
- bitchat mesh tops France's App Store, outranking Claude, Gemini and ChatGPT — TinfoilTricorn · 2026-10-24